← Back
ICRA 2026
Lateral Reciprocal Collision Avoidance: A Probabilistic Social-Norm-Inspired Strategy for Deadlock-Free Multi-Robot Navigation
Siyi Lu, Sipu Ruan
AI summary
Introducing randomized lateral displacement inspired by human social norms eliminates deadlock in decentralized multi-robot navigation while preserving efficiency.
Problem
Decentralized reciprocal collision avoidance methods frequently suffer from deadlock in symmetric or dense environments, severely degrading navigation success and efficiency.
Approach
The proposed LRCA framework embeds a probabilistic lateral avoidance norm into velocity obstacle constraints, using randomization to break symmetry and guide robots past each other without stopping.
Key results
- Formalized RCA deadlock via VO-based QPs and KKT conditions
- Proposed LRCA, a deadlock-free strategy inspired by pedestrian lateral avoidance
- Achieved near 100% success rate in symmetric and dense scenarios where ORCA fails
- Maintained ORCA-level computational efficiency while improving navigation metrics
Why it matters
Provides a scalable, deadlock-free solution for real-world multi-robot systems operating in dense or symmetric environments without heavy computational overhead.
Abstract
No abstract on file.